Printing the Underdetermined: Materializing Multi-solutionness in Figurative Paintings

arXiv cs.GR details a workflow that treats figurative paintings as inherently underdetermined rather than as single-scene puzzles. Instead of collapsing the image into one “correct” 3D model, the method preserves multiple compatible interpretations and makes that ambiguity visible in both digital and physical form.

The pipeline samples several camera-orbit multi-view video sequences from one painting, then reconstructs each sequence with 3D Gaussian Splatting. Because the added views come from a generative model without explicit 3D constraints, small frame-to-frame drift is expected, and the resulting Gaussian scenes can show density halos and ghosting that expose unresolved geometry.

The practical twist is DreamPrinting, which fabricates those reconstructions as physical artifacts. That turns uncertainty into something artists, researchers, and technical teams can inspect side by side, rather than treating it as an error to be eliminated.

For game developers, the interesting angle is less about painting analysis and more about representation: when input is ambiguous, a pipeline can preserve multiple valid spatial readings instead of overfitting to one. That has obvious resonance for concept art workflows, reconstruction tools, and any system that has to infer 3D from sparse or contradictory visual cues.